ztrans_proposal.mdwn

   1 # Ztrans - transcendental operations
   2
   3 See:
   4
   5 * <http://bugs.libre-riscv.org/show_bug.cgi?id=127>
   6 * <https://www.khronos.org/registry/spir-v/specs/unified1/OpenCL.ExtendedInstructionSet.100.html>
   7 * [[rv_major_opcode_1010011]] for opcode listing.
   8
   9 Extension subsets:
  10
  11 * **Ztrans**: standard transcendentals (best suited to 3D)
  12 * **ZtransExt**: extra functions (useful, not generally needed for 3D, can be synthesised using Ztrans)
  13 * **Ztrigpi**: trig. xxx-pi sinpi cospi tanpi
  14 * **Ztrignpi**: trig non-xxx-pi sin cos tan
  15 * **Zarctrigpi**: arc-trig. a-xxx-pi: atan2pi asinpi acospi
  16 * **Zarctrignpi**: arc-trig. non-a-xxx-pi: atan2, asin, acos
  17 * **Ztrignh**: trig/arc-trig hyperbolic.  sinh cosh tanh, asinh, acosh, atanh
  18 * **ZtransAdv**: much more complex to implement in hardware
  19
  20 Minimum recommended requirements for 3D: Ztrans, Ztrigpi, Zarctrigpi,
  21 Zarctrignpi
  22
  23 [[!toc levels=2]]
  24
  25 # List of 2-arg opcodes
  26
  27 [[!table  data="""
  28 opcode    | Description           | pseudo-code                | Extension |
  29 FATAN2    | atan2 arc tangent     | rd = atan2(rs2, rs1)       | Zarctrignpi |
  30 FATAN2PI  | atan arc tangent / pi | rd = atan2(rs2, rs1) / pi  | Zarctrigpi |
  31 FPOW      | x power of y          | rd = pow(rs1, rs2)         | ZtransAdv |
  32 FROOT     | x power 1/y           | rd = pow(rs1, 1/rs2)       | ZtransAdv |
  33 """]]
  34
  35 # List of 1-arg transcendental opcodes
  36
  37 [[!table  data="""
  38 opcode   | Description              | pseudo-code             | Extension |
  39 FCBRT    | Cube Root                | rd = pow(rs1, 3)        | Ztrans    |
  40 FEXP2    | power-of-2               | rd = pow(2, rs1)        | Ztrans    |
  41 FLOG2    | log2                     | rd = log2(rs1)          | Ztrans    |
  42 FEXPM1   | exponent minus 1         | rd = pow(e, rs1) - 1.0  | Ztrans    |
  43 FLOG1P   | log plus 1               | rd = log(e, 1 + rs1)    | Ztrans    |
  44 FEXP     | exponent                 | rd = pow(e, rs1)        | ZtransExt |
  45 FLOG     | natural log (base e)     | rd = log(e, rs1)        | ZtransExt |
  46 FEXP10   | power-of-10              | rd = pow(10, rs1)       | ZtransExt |
  47 FLOG10   | log base 10              | rd = log10(rs1)         | ZtransExt |
  48 """]]
  49
  50 # List of 1-arg trigonometric opcodes
  51
  52 [[!table  data="""
  53 opcode   | Description              | pseudo-code             | Extension |
  54 FSIN     | sin (radians)            | rd = sin(rs1)           | Ztrignpi    |
  55 FCOS     | cos (radians)            | rd = cos(rs1)           | Ztrignpi    |
  56 FTAN     | tan (radians)            | rd = tan(rs1)           | Ztrignpi    |
  57 FASIN    | arcsin (radians)         | rd = asin(rs1)          | Zarctrignpi |
  58 FACOS    | arccos (radians)         | rd = acos(rs1)          | Zarctrignpi |
  59 FSINPI   | sin times pi             | rd = sin(pi * rs1)      | Ztrigpi |
  60 FCOSPI   | cos times pi             | rd = cos(pi * rs1)      | Ztrigpi |
  61 FTANPI   | tan times pi             | rd = tan(pi * rs1)      | Ztrigpi |
  62 FASINPI  | arcsin times pi          | rd = asin(pi * rs1)     | Zarctrigpi |
  63 FACOSPI  | arccos times pi          | rd = acos(pi * rs1)     | Zarctrigpi |
  64 FATANPI  | arctan times pi          | rd = atan(pi * rs1)     | Zarctrigpi |
  65 FSINH    | hyperbolic sin (radians) | rd = sinh(rs1)          | Ztrigh |
  66 FCOSH    | hyperbolic cos (radians) | rd = cosh(rs1)          | Ztrigh |
  67 FTANH    | hyperbolic tan (radians) | rd = tanh(rs1)          | Ztrigh |
  68 FASINH   | inverse hyperbolic sin   | rd = asinh(rs1)         | Ztrigh |
  69 FACOSH   | inverse hyperbolic cos   | rd = acosh(rs1)         | Ztrigh |
  70 FATANH   | inverse hyperbolic tan   | rd = atanh(rs1)         | Ztrigh |
  71 """]]
  72
  73 # Pseudo-code ops and macro-ops
  74
  75 The pseudo-ops are best left up to the compiler rather than being actual
  76 pseudo-ops, by allocating one scalar FP register for use as a constant
  77 (loop invariant) set to "1.0" at the beginning of a function or other
  78 suitable code block.
  79
  80 * FRCP rd, rs1 - pseudo-code alias for rd = 1.0 / rs1
  81 * FATAN - pseudo-code alias for rd = atan2(rs1, 1.0) - FATAN2
  82 * FATANPI - pseudo alias for rd = atan2pi(rs1, 1.0) - FATAN2PI
  83 * FSINCOS - fused macro-op between FSIN and FCOS (issued in that order).
  84 * FSINCOSPI - fused macro-op between FSINPI and FCOSPI (issued in that order).
  85
  86 FATANPI example pseudo-code:
  87
  88     lui t0, 0x3F800 // upper bits of f32 1.0
  89     fmv.x.s ft0, t0
  90     fatan2pi.s rd, rs1, ft0
  91